Brain-Computer Interfaces Are Leaving the Lab — What Comes Next?

In January 2024, a 29-year-old paralyzed Arizona man named Noland Arbaugh became the first human patient to receive an N1 brain-computer interface implant from Neuralink. Within weeks of his surgery, video appeared online of Arbaugh playing chess on a laptop by thinking, then playing Civilization, then web browsing — controlling a cursor with no observable hand movement, no eye-tracking, no muscle twitches. The footage was striking enough that it briefly redefined the public’s mental model of what brain implants could do.

What it didn’t show — and what Neuralink later acknowledged — was that within a few months, several of the threads carrying electrodes into Arbaugh’s motor cortex had partially retracted, reducing the device’s recorded neural channel count. The team compensated with software updates, and Arbaugh kept using the implant. But the story turned out to be more complicated than the chess clip suggested: a real human, a real implant, real capabilities — and a real engineering problem about how to keep tiny electrodes anchored in living brain tissue for years rather than months.

This is roughly the state of brain-computer interfaces in 2026. Years of laboratory progress have produced devices that genuinely work in real patients, in real homes, doing real tasks. Multiple companies are running FDA-approved trials. The hardware-and-biology problem of getting durable, high-bandwidth signals out of the human cortex is partly solved. What’s still mostly unsolved is the distance between first patient and medical product — and that gap, paradoxically, is where the next ten years of BCI work will happen.

Two decades in the lab

The current generation of commercial BCI startups stands on a foundation built almost entirely by one academic consortium: BrainGate.

Founded in the early 2000s as a collaboration among Brown University, Massachusetts General Hospital, Stanford, the VA, and several other research centers, BrainGate took the Utah Array — a tiny silicon block bristling with about 100 microelectrode needles, originally developed in the 1990s by Richard Normann at the University of Utah — and began implanting it in human volunteers with severe paralysis. The first BrainGate patient, Matthew Nagle, received an implant in 2004 and learned to control a computer cursor with his thoughts. He could open emails. The result was published in Nature in 2006.

What followed was a long, patient accumulation of capability demonstrations across a small population of volunteers. In 2012, BrainGate participant Cathy Hutchinson — paralyzed by a brainstem stroke — used the system to direct a robotic arm to pick up a bottle of coffee and drink from it. The video circulated widely and arguably became the founding moment of the modern BCI era. In 2016, a Pittsburgh patient named Nathan Copeland, working with researchers at the University of Pittsburgh, became the first person to receive tactile feedback through a BCI — pressure applied to a robotic arm registered as sensation in his fingers, even though his actual fingers were paralyzed.

Twenty years between the first human implant and the first patient who got to keep one outside a hospital.

Subsequent demonstrations kept pushing the bandwidth boundary. In 2023, a Stanford team led by Frank Willett and Krishna Shenoy reported that BrainGate participant Pat Bennett — who has ALS — could “type” via decoded speech at roughly 62 words per minute, with another patient at UCSF (Edward Chang’s lab, separately funded) achieving similar rates through a different cortical-surface approach. Twenty years after Nagle’s cursor demo, paralyzed patients were communicating via brain implants at speeds that approached the lower end of natural conversation.

But every one of those patients was, until very recently, a research subject — implanted under an academic IRB protocol, in a study, in a hospital. The implants were never sold as products. None of them went home with a device they personally owned.

The current commercial pipeline

What changed between 2020 and 2024 is the emergence of companies trying to take that science out of academic centers and into the FDA’s medical-device pipeline. Each is pursuing a meaningfully different architectural bet.

Neuralink (founded 2016 in Fremont, California) implants a thread-based array — about 64 flexible polymer threads, each carrying multiple electrodes, robotically inserted into specific cortical regions. The N1 device is fully implanted in the skull with no external connectors, charging wirelessly through the scalp. Arbaugh’s January 2024 surgery was the first under FDA-approved human trial; subsequent patients have followed. Neuralink’s bet is on the highest electrode count of any human BCI to date and on a closed cosmetic profile that doesn’t require percutaneous hardware.

Synchron (CEO Tom Oxley, with offices in Brooklyn and Melbourne) takes the opposite architectural approach: instead of implanting through the skull into cortical tissue, its Stentrode is a stent-like device threaded through the jugular vein and parked inside a vein that runs across the motor cortex. Electrodes on the stent record signals through the vessel wall. The procedure is endovascular — closer to placing a heart stent than to brain surgery — and the device became the first BCI to receive an FDA Investigational Device Exemption, in 2021. Synchron has implanted around ten patients in the COMMAND trial in the U.S. and Australia, with subjects using the system for tasks like texting and emailing.

Onward Medical (Eindhoven, Netherlands) isn’t strictly a BCI company in the cortical sense — its ARC therapy combines an implanted spinal cord stimulator with brain signals to restore motor control in paralyzed patients. The company became internationally famous in 2023 when its collaboration with Grégoire Courtine and Jocelyne Bloch’s lab at EPFL (Lausanne) restored the ability to walk to Gert-Jan Oskam, a Dutch man paralyzed in a cycling accident. Oskam’s brain-spine interface — a cortical implant that decodes movement intent, paired with a spinal stimulator that produces the corresponding muscle activation — is the closest thing yet to bypassing a damaged spinal cord with electronics.

Blackrock Neurotech (Salt Lake City) is the commercial successor to the original Utah Array work and continues to supply the hardware most BrainGate patients have used for two decades. Its MoveAgain BCI is in late-stage development; the company’s strategy emphasizes building on the longest accumulated safety record of any cortical implant in humans.

Precision Neuroscience (founded by ex-Neuralink co-founder Benjamin Rapoport) builds a thin-film cortical surface array — the Layer 7 Cortical Interface — that sits on top of the brain rather than penetrating it. The bet is that surface electrodes, with very high density, can deliver enough signal for most BCI applications without the long-term tissue-response problems that come with penetrating arrays.

Paradromics (Austin, Texas) is pursuing the highest-bandwidth penetrating BCI in development, with its Connexus device targeting tens of thousands of channels — roughly two orders of magnitude beyond current human implants. Its first human trials were authorized in 2024.

The brain doesn’t care which architecture you pick. It just wants the signal back.

Three architectural bets

The companies above are placing distinct bets on what shape BCIs will eventually take, and the differences matter.

The penetrating-array bet (Neuralink, Blackrock, Paradromics) is that putting electrodes inside cortical tissue, where individual neurons fire millimeters away, gives the cleanest, highest-resolution signal — but pays the cost of biological foreign-body response, electrode encapsulation by glial scar tissue, and gradual signal degradation over months to years. Solving that durability problem is the central engineering challenge.

The endovascular bet (Synchron) is that you can get useful signal — not the cleanest, but adequate for cursor control and communication — without ever cutting into the brain. The blood vessel acts as a lower-resolution interface, but the safety and accessibility profile is dramatically better. If Synchron is right that the bandwidth is enough for clinically useful tasks, its device could reach orders of magnitude more patients than any penetrating system.

The surface bet (Precision) splits the difference: electrodes touching but not piercing the cortex, with very high spatial density, ideally avoiding the worst foreign-body responses while keeping reasonable signal quality. It’s the youngest of the three approaches with the least clinical track record, but the rationale is strong if it pans out.

None of these bets is obviously correct, and they may not be mutually exclusive. Different patient populations and different applications could end up matching different architectures.

What “leaving the lab” actually means

The hard part is no longer the cortical signal. It’s the rest of the medical-device system around it.

Real BCI products need to clear FDA Premarket Approval — the most stringent regulatory pathway in U.S. medicine, reserved for high-risk Class III devices. They need supply chains for implant manufacturing at consistent quality. They need surgical training programs for the neurosurgeons who’ll do the implants. They need long-term reliability data — which, by definition, requires waiting years to collect. They need insurance coverage codes, because no one is paying $100K-$500K out of pocket for a brain implant. They need software ecosystems robust enough that the device works in a patient’s home for a decade without an engineer flying out every time something glitches.

Each of those is its own multi-year problem. Insurance coverage, in particular, is a quietly enormous bottleneck — the Medicare and private-payer process for new high-cost devices typically lags FDA approval by years. The patients with the most acute need (people with ALS, severe spinal cord injuries, locked-in syndrome) are also disproportionately reliant on Medicare and Medicaid, which means coverage decisions matter as much as approval decisions.

What’s next

A reasonable expectation for the rest of this decade: at least one and probably two BCI devices will receive full FDA clearance for clinical use, likely first for narrow indications like restoring communication for ALS or severe paralysis. Patient populations will be small — measured in hundreds, not thousands — and access will be concentrated in major academic medical centers. The hype-versus-reality gap will keep being uncomfortable: clips of patients controlling cursors will continue to circulate while the actual rollout proceeds at the speed of medicine, not the speed of consumer technology.

What was unimaginable in 2002 is FDA-pathway in 2024. What’s FDA-pathway in 2024 will be standard of care in 2034.

Underneath that, the more interesting story is the one that doesn’t make video clips: the gradual standardization of cortical implants from one-off research curiosities into reproducible, manufacturable, regulable medical devices. What was unimaginable in 2002 is FDA-pathway in 2024. What’s FDA-pathway in 2024 will be standard of care in 2034 — for a small but real population of patients whose lives depend on it.

The lab era of BCI is genuinely ending. The medical-device era is beginning, and that era’s protagonists are insurance reviewers, manufacturing engineers, and FDA panel members at least as much as neuroscientists.